Cognitive and Semantic Prototype–Based Decision–Making for Game Agents

• Augusto Baffa PUC-Rio
• Edward Hermann PUC-Rio

DOI: https://doi.org/10.5753/sbgames.2026.25542

Resumo

Introduction: Decision–making in games involves a trade–off between interpretability and adaptability. Symbolic approaches such as Finite State Machines (FSMs) provide transparency but limited flexibility, while deep reinforcement learning methods such as Proximal Policy Optimization (PPO) offer performance, but need extensive training data and are harder to explain. Objective: This work investigates prototype–based decision–making and how it can serve as a middle ground between symbolic and sub–symbolic methods, combining interpretability and adaptive performance in real–time environments. Methodology: A cognitive architecture based on prototype theory is proposed, integrating semantic state representation, conceptual memory, and local reinforcement through multi–armed bandits, without neural networks. A key feature is the separation between perceptual similarity and decision competence. The approach is evaluated in an Asteroids environment against FSM and PPO baselines under identical conditions. Results: The proposed controller achieved overall performance in terms of return, score, and stability relative to the FSM and PPO baselines under identical test conditions. This suggests that prototype–based decision–making may provide an interpretable alternative with adaptive performance comparable to the baselines.

Palavras-chave: Game AI, Interpretable Reinforcement Learning, Prototype–Based Decision Making, Multi–Armed Bandits, Cognitive Architectures

Referências

Anderson, J. R. (1990). The Adaptive Character of Thought. Lawrence Erlbaum.

Auer, P., Cesa-Bianchi, N., e Fischer, P. (2002). Finite-time analysis of the multiarmed bandit problem. Machine Learning, 47:235–256.

Bakkes, S. C., Spronck, P. H., e van Lankveld, G. (2012). Player behavioural modelling for video games. Entertainment Computing, 3(3):71–79. Games and AI.

Bazzaz, M. e Cooper, S. (2024). Guided game level repair via explainable ai. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 20, pages 139–148.

Bekkemoen, Y. e Langseth, H. (2025). Interpretable Deep Reinforcement Learning Via Concept-Based Policy Distillation. Machine Learning, 114(12):288.

Cai, M., Wei, G., e Cao, J. (2019). Categories in emergency decision-making: prototype-based classification. Kybernetes, 49(2):526–553. _eprint: [link].

Chaudhuri, A. R., Jawanpuria, P., e Mishra, B. (2023). ProtoBandit: Efficient prototype selection via multi-armed bandits. In Khan, E. e Gonen, M., editors, Proceedings of The 14th Asian Conference on Machine Learning, volume 189 of Proceedings of Machine Learning Research, pages 169–184. PMLR.

Chen, C., Li, O., Tao, D., Barnett, A., Rudin, C., e Su, J. K. (2019). This looks like that: deep learning for interpretable image recognition. volume 32.

Doshi-Velez, F. e Kim, B. (2017). Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608.

Douven, I., Verheyen, S., Elqayam, S., Gärdenfors, P., e Osta-Vélez, M. (2023). Similarity-based reasoning in conceptual spaces. Frontiers in Psychology, 14:1234483.

d’Avila Garcez, A. S., Lamb, L. C., e Gabbay, D. M. (2009). Neural-symbolic cognitive reasoning. Springer.

Gärdenfors, P. (2000). Conceptual spaces: The geometry of thought. A Bradford book, volume 3.

Glanois, C., Weng, P., Zimmer, M., Li, D., Yang, T., Hao, J., e Liu, W. (2024). A survey on interpretable reinforcement learning. Machine Learning, 113(8):5847–5890.

Hessel, M., Modayil, J., Van Hasselt, H., Schaul, T., Ostrovski, G., Dabney, W., Horgan, D., Piot, B., Azar, M., e Silver, D. (2018). Rainbow: Combining improvements in deep reinforcement learning. In Proceedings of the AAAI conference on artificial intelligence, volume 32.

Kenny, E. M., Tucker, M., e Shah, J. (2023). Towards interpretable deep reinforcement learning with human-friendly prototypes. In The Eleventh International Conference on Learning Representations.

Li, P., Siddique, U., e Cao, Y. (2025). From explainability to interpretability: Interpretable reinforcement learning via model explanations. In Reinforcement Learning Conference.

Milani, S., Topin, N., Veloso, M., e Fang, F. (2024). Explainable reinforcement learning: A survey and comparative review. ACM Computing Surveys, 56(7):1–36.

Millington, I. e Funge, J. (2009). Artificial Intelligence for Games. Morgan Kaufmann.

Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M. A., Fidjeland, A. K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S., e Hassabis, D. (2015). Human-level control through deep reinforcement learning. Nature, 518:529– 533.

Newell, A. (1994). Unified theories of cognition. Harvard University Press.

Proietti, M., Wurman, P. R., Stone, P., e Capobianco, R. (2025). Protocrl: Prototype-based network for continual reinforcement learning.

Rosch, E. (1975). Cognitive representations of semantic categories. Journal of experimental psychology: General, 104(3):192.

Samuel, B., Treanor, M., e McCoy, J. (2021). Design considerations for creating ai-based gameplay. In AIIDE Workshops.

Schulman, J., Wolski, F., Dhariwal, P., Radford, A., e Klimov, O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347.

Sun, S., Qi, M., e Shen, Z.-J. M. (2025). Online mdp with prototypes information: A robust adaptive approach. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 20717–20724.

Tappler, M., Lopez-Miguel, I. D., Tschiatschek, S., e Bartocci, E. (2025). Rule-Guided Reinforcement Learning Policy Evaluation and Improvement. In Kwok, J., editor, Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, pages 6254–6262. International Joint Conferences on Artificial Intelligence Organization.

Vamshi, B. K. e Yang, H. (2026). Principal prototype analysis on manifold for interpretable reinforcement learning. arXiv preprint arXiv:2603.27971.

Verma, A., Murali, V., Singh, R., Kohli, P., e Chaudhuri, S. (2018). Programmatically interpretable reinforcement learning. In International conference on machine learning, pages 5045–5054. PMLR.